python3 test failures

[juliantaylor]

on ubuntu 13.04 several tests fail with 3.3 while they suceed with python2
13.04 has python 3.3 and numpy 1.7
current pandas head 61766ec

======================================================================
FAIL: test_quoting (pandas.io.tests.test_parsers.TestParsers)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/home/jtaylor/tmp/pandas-0.9.1/build/lib.linux-x86_64-3.3/pandas/io/tests/test_parsers.py", line 528, in test_quoting
    sep='\t')
AssertionError: Exception not raised by read_table

======================================================================
FAIL: test_cant_compare_tz_naive_w_aware (pandas.tseries.tests.test_timeseries.TestTimestamp)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/home/jtaylor/tmp/pandas-0.9.1/build/lib.linux-x86_64-3.3/pandas/tseries/tests/test_timeseries.py", line 2349, in test_cant_compare_tz_naive_w_aware
    self.assertRaises(Exception, a.__eq__, b.to_pydatetime())
AssertionError: Exception not raised by __eq__

----------------------------------------------------------------------

======================================================================
FAIL: test_more_flexible_frame_multi_function (__main__.TestGroupBy)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/home/jtaylor/tmp/pandas-0.9.1/build/lib.linux-x86_64-3.3/pandas/tests/test_groupby.py", line 1909, in test_more_flexible_frame_multi_function
    assert_frame_equal(result, expected)
  File "/home/jtaylor/tmp/pandas-0.9.1/build/lib.linux-x86_64-3.3/pandas/util/testing.py", line 167, in assert_frame_equal
    assert(left.columns.equals(right.columns))
AssertionError

----------------------------------------------------------------------

======================================================================
ERROR: test_yahoo (pandas.io.tests.test_yahoo.TestYahoo)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/tmp/pandas/build/lib.linux-x86_64-3.3/pandas/io/tests/test_yahoo.py", line 25, in test_yahoo
    pd.DataReader("F", 'yahoo', start, end)['Close'][-1],
  File "/tmp/pandas/build/lib.linux-x86_64-3.3/pandas/io/data.py", line 58, in DataReader
    retry_count=retry_count, pause=pause)
  File "/tmp/pandas/build/lib.linux-x86_64-3.3/pandas/io/data.py", line 149, in get_data_yahoo
    parse_dates=True)[::-1]
  File "/tmp/pandas/build/lib.linux-x86_64-3.3/pandas/io/parsers.py", line 364, in parser_f
    return _read(filepath_or_buffer, kwds)
  File "/tmp/pandas/build/lib.linux-x86_64-3.3/pandas/io/parsers.py", line 195, in _read
    return parser.read()
  File "/tmp/pandas/build/lib.linux-x86_64-3.3/pandas/io/parsers.py", line 592, in read
    ret = self._engine.read(nrows)
  File "/tmp/pandas/build/lib.linux-x86_64-3.3/pandas/io/parsers.py", line 844, in read
    data = self._reader.read(nrows)
  File "parser.pyx", line 597, in pandas._parser.TextReader.read (pandas/src/parser.c:5342)
  File "parser.pyx", line 619, in pandas._parser.TextReader._read_low_memory (pandas/src/parser.c:5562)
  File "parser.pyx", line 668, in pandas._parser.TextReader._read_rows (pandas/src/parser.c:6143)
  File "parser.pyx", line 655, in pandas._parser.TextReader._tokenize_rows (pandas/src/parser.c:6027)
  File "parser.pyx", line 1385, in pandas._parser.raise_parser_error (pandas/src/parser.c:14807)
pandas._parser.CParserError: Error tokenizing data. C error: Expected 7 fields in line 106, saw 3

test_more_flexible_frame_multi_function (__main__.TestGroupBy) ... > /home/jtaylor/tmp/pandas-0.9.1/build/lib.linux-x86_64-3.3/pandas/util/testing.py(167)assert_frame_equal()
-> assert(left.columns.equals(right.columns))
(Pdb) p left.columns

[(D, mean), (D, std), (C, mean), (C, std)]
(Pdb) p right.columns
MultiIndex
[(C, mean), (C, std), (D, mean), (D, std)]

this one is random probably related to hash randomization

the yahoo one might be due to missing internet connection?

[juliantaylor]

another one with matplotlib installed

======================================================================
ERROR: test_bootstrap_plot (pandas.tests.test_graphics.TestSeriesPlots)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/usr/lib/python3/dist-packages/pandas/tests/test_graphics.py", line 158, in test_bootstrap_plot
    _check_plot_works(bootstrap_plot, self.ts, size=10)
  File "/usr/lib/python3/dist-packages/pandas/tests/test_graphics.py", line 625, in _check_plot_works
    ret = f(*args, **kwargs)
  File "/usr/lib/python3/dist-packages/pandas/tools/plotting.py", line 376, in bootstrap_plot
    samplings = [random.sample(data, size) for _ in range(samples)]
  File "/usr/lib/python3/dist-packages/pandas/tools/plotting.py", line 376, in <listcomp>
    samplings = [random.sample(data, size) for _ in range(samples)]
  File "/usr/lib/python3.3/random.py", line 298, in sample
    raise TypeError("Population must be a sequence or set.  For dicts, use list(d).")
TypeError: Population must be a sequence or set.  For dicts, use list(d).

seems like random.sample does not take np arrays, its no abc.Sequence (which might be a bug in numpy?)

[ghost]

I can reproduce Intermittent failures on py3.3 with numpy 1.7.0rc1.dev-3a52aa0' (git master).
I cannot reproduce failures on py3.2 with numpy 1.7.
I cannot reproduce failures on py2.7 with numpy 1.7.

test_dict sometimes also fails.

[juliantaylor]

the random failures can only be seen with python 3.3 and python 2.7.3, hash randomization does not exist earlier
with python2.7.3 it might have to explicitly enable it

[ghost]

verified. can reproduce on 2.7.3 with hash randomization enabled.
here's the announcement.
The change mitigates an algorithmic complexity attack.

[juliantaylor]

actually no, the python "fix" is completely ineffective (see issue14621)
nevertheless relying on ordering of dicts was never supported an should thus be fixed

[ghost]

afaict, the seed is only randomized between sessions, traversal order within a given process is
determinstic.
Of course relying on order should be fixed, but I'm not sure why the problem is surfacing,
I don't see any state being saved across runs.

[ghost]

dear lord, the order of keys is this form of construction is not well-determined:

OrderedDict(foo=0,bar=1)

[wesm]

Python 3.3 with NumPy git master passing cleanly for me now. Could you give it a shot with your environment and let me know so I can close this issue?

[juliantaylor]

I still get this failure (f4b5cbb):

======================================================================
FAIL: test_quoting (pandas.io.tests.test_parsers.TestPythonParser)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/pandas/build/lib.linux-x86_64-3.3/pandas/io/tests/test_parsers.py", line 462, in test_quoting
    sep='\t')
AssertionError: Exception not raised by read_table

----------------------------------------------------------------------

[juliantaylor]

that one only fails with LC_ALL=C
it works with C.UTF-8

edit, not it fails with utf-8 too
it also fails in python2 for me now too

[wesm]

I tried doing LC_ALL=C; nosetests pandas/io/tests/test_parsers.py and it worked-- I'm on 64-bit Ubuntu 11.04, maybe that's not the right way?

[juliantaylor]

I can only reproduce it in ubuntu 13.04, not 12.10. both use python 2.7.3
so far I know that uses a snapshot of pythons stable 2.7.3 branch, maybe some change causes the issue

[wesm]

Bummer. I'll close this issue until it becomes a problem elsewhere-- the test case that is failing is very esoteric, I wouldn't expect you to run into it in the course of normal use. If you get motivated and want to pick through the C file tokenizer and figure out why it isn't falling into the error state, be my guest.

[ghost]

@wesm, i get the same failure consistently on 64 bit debian testing with 3.3.

[juliantaylor]

I confirmed it works with ubuntus 2.7.3-5ubuntu4 but not 2.7.3-5ubuntu5
the change log us unfortunately long:

• Update to 20121128, taken from the 2.7 branch.
  • Issue TimedeltaIndex constructor gives wild output #9011: Fix hacky AST code that modified the CST when compiling
    a negated numeric literal.
  • Issue some warnings not being silenced #16306: Fix multiple error messages when unknown command line
    parameters where passed to the interpreter.
  • Issue Compat with jupyter's new HTML table style #15379: Fix passing of non-BMP characters as integers for the
    charmap decoder (already working as unicode strings).
  • Issue BUG: Index values not read-only #16453: Fix equality testing of dead weakref objects.
  • Issue read_csv(<file>, nrows=x) raises an StopIteration exception when given file doesn't contain any newlines #9535: Fix pending signals that have been received but not yet
    handled by Python to not persist after os.fork() in the child process.
  • Issue CLN/DEPR: Consistent naming for isnull, notnull, fillna, dropna #15001: fix segfault on "del sys.modules['main']".
  • Issue DOC: proposed fix for #4699: Period() docstring inconsistent with code when freq has a mult != 1 #5057: the peepholer no longer optimizes subscription on unicode
    literals (e.g. u'foo'[0]) in order to produce compatible pyc files
    between narrow and wide builds.
  • Issue BUG: Panel.fillna with method='ffill' ignores the axis parameter (GH8251) #8401: assigning an int to a bytearray slice (e.g. b[3:4] = 5)
    now raises an error.
  • Issue CLN: remove bare excepts #14700: Fix buggy overflow checks for large width and precision
    in string formatting operations.
  • Issue result from groupby / nlargest with data frame with one row does not include the groupby key in the resulting index #16345: Fix an infinite loop when fromkeys on a dict subclass
    received a nonempty dict from the constructor.
  • Issue TST: Fix test_eval.py #6074: Ensure cached bytecode files can always be updated by the
    user that created them, even when the source file is read-only.
  • Issue ENH: merge_asof() has type specializations and can take multiple 'by' parameters (#13936) #14783: Improve int() and long() docstrings and switch docstrings
    for unicode(), slice(), range(), and xrange() to use multi-line
    signatures.
  • Issue TST: reduce amount of nesting in tests, specifically move core routines higher #16030: Fix overflow bug in computing the repr of an xrange
    object with large start, step or length.
  • Issue TST: partial indexing with __getitem__ and integer labels #16029: Fix overflow bug occurring when pickling xranges with large
    start, step or length.
  • Issue ENH: support .groupby().ewm() directly #16037: Limit httplib's _read_status() function to work around
    broken HTTP servers and reduce memory usage (backport of a 3.2 fix).
  • Issue SeriesGroupby.cumsum raises on object dtype #13992: The trashcan mechanism is now thread-safe. This eliminates
    sporadic crashes in multi-thread programs when several long deallocator
    chains ran concurrently and involved subclasses of built-in container
    types.
  • Issue TST/BUG: HDFStore.append_to_multiple with dropna=False buggy #15801: Make sure mappings passed to '%' formatting are actually
    subscriptable.
  • Issue pandas.Series.sort_values(ascending=[False]) behaves as ascending=True #15604: Update uses of PyObject_IsTrue() to check for and handle
    errors correctly.
  • Issue DEPR: correct locations to access public tslib objects #15897: zipimport.c doesn't check return value of fseek().
  • Issue TST: Add test for clip-na #16369: Global PyTypeObjects not initialized with PyType_Ready(...).
  • Issue Enhancement: add "median" to series.describe() #15033: Fix the exit status bug when modules invoked using
    -m switch, return the proper failure return value (1).
  • Issue Should avoid hash table creation in monotonic DatetimeIndex? #1160: Fix compiling large regular expressions on UCS2 builds.
  • Issue BUG: resample and apply modify the index type for empty Series #14313: zipfile now raises NotImplementedError when the compression
    type is unknown.
  • Issue API: Infer dtype scalar #16408: Fix file descriptors not being closed in error conditions
    in the zipfile module.
  • Issue Plotting time series returns ImportError: cannot import name 'checked_add_with_arr' #16327: The subprocess module no longer leaks file descriptors
    used for stdin/stdout/stderr pipes to the child when fork() fails.
  • Issue pandas.*.sample may return empty objects if frac <= 2^{-axis_length} #14396: Handle the odd rare case of waitpid returning 0 when not
    expected in subprocess.Popen.wait().
  • Issue Update tutorials.rst #16411: Fix a bug where zlib.decompressobj().flush() might try
    to access previously-freed memory.
  • Issue DOC: dev docs authentication failing #16350: zlib.decompressobj().decompress() now accumulates data from
    successive calls after EOF in unused_data, instead of only saving the
    argument to the last call. decompressobj().flush() now correctly sets
    unused_data and unconsumed_tail. A bug in the handling of MemoryError
    when setting the unconsumed_tail attribute has also been fixed.
  • Issue API: Change Period('NAT') to return NaT, not Period instance #12759: sre_parse now raises a proper error when the name
    of the group is missing.
  • Issue CLN: remove unused TimeGrouper._get_binner_for_resample() method #16152: fix tokenize to ignore whitespace at the end of the code
    when no newline is found.
  • Issue #1207589: Add Cut/Copy/Paste items to IDLE right click
    Context Menu.
  • Issue BUG: concat of tz-aware with missing #16230: Fix a crash in select.select() when one the lists changes
    size while iterated on.
  • Issue ERR: left_index AND left_on or right_index AND right_on should raise #16228: Fix a crash in the json module where a list changes size
    while it is being encoded.
  • Issue DOC : build isnt a valid argument to doc/make.py #14897: Enhance error messages of struct.pack and struct.pack_into.
  • Issue The API to retrieve serie elements presents some inconsistencies #12890: cgitb no longer prints spurious

    tags in text
    mode when the logdir option is specified.

  • Issue Incompatibility between pandas.infer_freq and pandas.to_timedelta #14398: Fix size truncation and overflow bugs in the bz2 module.
  • Issue ENH: Cleanup backend for Offsets and Period #5148: Ignore 'U' in mode given to gzip.open() and gzip.GzipFile().
  • Issue ENH: Provide dict object for to_dict() #16122 #16220: wsgiref now always calls close() on an iterable response.
  • Issue BUG: strange timeseries plot behavior #16461: Wave library should be able to deal with 4GB wav files,
    and sample rate of 44100 Hz.
  • Issue MAINT: Remove self.assertNotEqual from testing #16176: Properly identify Windows 8 via platform.platform().
  • Issue ENH: support "nrows" and "chunksize" together #15756: subprocess.poll() now properly handles errno.ECHILD to
    return a returncode of 0 when the child has already exited or cannot
    be waited on.
  • Issue Fix #12373: rolling functions raise ValueError on float32 data #12376: Pass on parameters in TextTestResult.init super call.
  • Issue DOC: Merge FAQ and gotcha (rebase of GH13768) #15222: Insert blank line after each message in mbox mailboxes.
  • Issue CLN: move pandas.formats -> pandas.io.formats #16013: Fix CSV Reader parsing issue with ending quote characters.
  • Issue Unexpected string->float conversion in DataFrame.groupby().apply() #15421: fix an OverflowError in Calendar.itermonthdates() after
    datetime.MAXYEAR.
  • Issue diff breaks when float64 column cast to float32 #15970: xml.etree.ElementTree now serializes correctly the
    empty HTML elements 'meta' and 'param'.
  • Issue BUG: merge_asof only merges by first column of 'by=' list #15676: Now "mmap" check for empty files before doing the
    offset check.
  • Issue BUG: .groupby on Series' values without reindexing #15340: Fix importing the random module when /dev/urandom cannot
    be opened. This was a regression caused by the hash randomization patch.
  • Issue CLN: spin msgpack support to [new] pandas-msgpack #15841: The readable(), writable() and seekable() methods of
    io.BytesIO and io.StringIO objects now raise ValueError when the object
    has been closed.
  • Issue SparseDataFrame.fillna() doesn't fill all NaNs #16112: platform.architecture does not correctly escape argument to
    /usr/bin/file.
  • Issue BUG: filter (with dropna=False) when there are no groups fulfilling the condition #12776,"NA" value from Excel read by Pandas as nan #11839: call argparse type function (specified by
    add_argument) only once. Before, the type function was called twice in
    the case where the default was specified and the argument was given as
    well. This was especially problematic for the FileType type, as a
    default file would always be opened, even if a file argument was
    specified on the command line.
  • Issue Slicing Datetime MultiIndex with string or datetime.date slices #15906: Fix a regression in argparse caused by the preceding change,
    when action='append', type='str' and default=[].
  • Issue Dataframes ready to be imported? #13370: Ensure that ctypes works on Mac OS X when Python is
    compiled using the clang compiler
  • Issue BUG: syntax error in hdf query with ts #15544: Fix Decimal.float to work with payload-carrying NaNs.
  • Issue BLD: numpy-dev causing some issues #15199: Fix JavaScript's default MIME type to
    application/javascript.
  • Issue CLN: split off frozen (immutable) data structures into pandas/indexes/frozen.py #15477: In cmath and math modules, add workaround for platforms
    whose system-supplied log1p function doesn't respect signs of zeros.
  • Issue Class label colours with parallel coordinates #15908: Fix misbehaviour of the sha1 module when called on data
    larger than 2**32 bytes.
  • Issue BUG: read_csv skipfooter fails with invalid quoted line #15910: Fix misbehaviour of _md5 and sha1 modules when "updating"
    on data larger than 2**32 bytes.
  • Fix the leak of a dict in the time module when used in an embedded
    interpreter that is repeatedly initialized and shutdown and reinitialized.
  • Issue added pd as namespace #12268: File readline, readlines and read or readall methods
    no longer lose data when an underlying read system call is interrupted
    within an io module object. IOError is no longer raised due to a read
    system call returning EINTR from within these methods.
  • Issue .isin() raises TypeError for Series with > 1 million elements if argument has mixed dtypes #16012: Fix a regression in pyexpat. The parser's UseForeignDTD()
    method doesn't require an argument again.
  • Issue Old read_csv() & EOF character issue back #16559: Add more tests for the json module, including some from the
    official test suite at json.org.
  • Issue DOC: change 0.20.1 whatsnew text -> 0.20.2 #16274: Fix test_asyncore on Solaris.
  • Issue DOC: Typing error in ExcelFile class example #15040: Close files in mailbox tests for PyPy compatibility.
  • Issue Revert "MAINT: Remove Long and WidePanel (#15748)" #15802: Fix test logic in TestMaildir.test_create_tmp.
  • Issue CLN: replace _interleave_dtype with _find_common_type #15765: Extend a previous fix to Solaris and OpenBSD for quirky
    getcwd() behaviour (issue API: restore timedelta compability for Timedelta components? #9185) to NetBSD as well.
  • Issue Clean multiindex keys #15615: Add some tests for the json module's handling of invalid
    input data.
  • Issue BUG pd.get_dummies does not propagate NaN correctly #15923: fix a mistake in asdl_c.py that resulted in a TypeError
    after 2801bf875a24 (see TST/BUG: HDFStore.append_to_multiple with dropna=False buggy #15801).
  • Issue Fix to BUG GH11698. added default value of mask #11715: Fix multiarch detection without having Debian development
    tools (dpkg-dev) installed.
  • Issue ERR: consistency in rolling on empty frames #15819: Make sure we can build Python out-of-tree from a readonly
    source directory. (Somewhat related to Issue Problems reading an specific attribute with a dataframe #9860.)
  • Issue get localized Timestamp using ".at" cannot get the time zone info #15822: Ensure 2to3 grammar pickles are properly installed.
  • Issue TST: reorg datetime with tz tests a bit #13301: use ast.literal_eval() instead of eval()
    in Tools/i18n/msgfmt.py.
  • Issue DOC: add section on Interval #16400: Update the description of which versions of a given package
    PyPI displays.
  • Issue API: df.rolling(..).corr()/cov() when pairwise=True to return MI DataFrame #15677: Document that zlib and gzip accept a compression level
    of 0 to mean 'no compression'.
  • Issue Pandas 0.14.1 StataReader seems to read .dta files 10x slower than Pandas 0.13.1 #8040: added a version switcher to the documentation.
  • Issue Regression in CategoricalIndex in 0.20rc1 #16115: Improve subprocess.Popen() documentation around args,
    shell, and executable arguments.
  • Issue Multi index operations #15979: Improve timeit documentation.
  • Issue read_csv on bzip2 file unzips only the first block #16036: Improve documentation of built-in int()'s signature and
    arguments.
  • Issue ENH: Minor change to parallel_coordinates (#15908) #15935: Clarification of argparse docs, re: add_argument() type and
    default arguments.
  • Issue TST: add timedelta describe #13769: Document the effect of ensure_ascii to the return type
    of JSON decoding functions.
  • Issue DOC for refactored compression (GH14576) + BUG: bz2-compressed URL with C engine (GH14874) #14880: Fix kwargs notation in csv.reader,
    .writer & .register_dialect.
  • Issue raise ValueError if usecols has same size as but doesn't exist in headers (#14671) #14674: Add a discussion of the json module's standard compliance.
• Clarify location of the gdbinit file. LP: #975676.
• Fix traceback for missing distutils wininst .exe files. LP: #1081155.
  -- Matthias Klose doko@ubuntu.com Wed, 28 Nov 2012 13:49:46 +0100

[wesm]

OK, I'll reopen.

[wesm]

@y-p if you have the stomach for it, you'll want to look at tokenize_delimited in pandas/src/parser/tokenizer.c. The error should be originating (I think?) because the state of the tokenizer is IN_QUOTED_FIELD when it reaches the end of the file, the result of a malformed line in the file. I bet if you change the string literal to a bytes literal it will fix the test. Otherwise have to dig

[ghost]

I will if I have to, but it looks like the failure is in the python rather then the c tokenizer:
I put print( locals()) at the start of _make_parser_function and got:

======================================================================
FAIL: test_quoting (pandas.io.tests.test_parsers.TestPythonParser)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/home/user1/src/pandas/.tox/py33/lib/python3.3/site-packages/pandas-0.10.0.dev_3950684-py3.3-linux-x86_64.egg/pandas/io/tests/test_parsers.py", line 462, in test_quoting
    sep='\t')
AssertionError: AssertionError: Exception not raised by read_table
-------------------- >> begin captured stdout << ---------------------
{'index_col': None, 'chunksize': None, 'true_values': None, 'skiprows': None, 'use_unsigned': False, 'na_values': None, 'keep_default_na': True, 'names': None, 'date_parser': None, 'buffer_lines': None, 'converters': None, 'as_recarray': False, 'parse_dates': False, 'quoting': 0, 'squeeze': False, 'memory_map': False, 'na_filter': True, 'skip_footer': 0, 'filepath_or_buffer': <_io.StringIO object at 0x7f52b6cc93e0>, 'skipinitialspace': False, 'low_memory': True, 'skipfooter': None, 'usecols': None, 'header': 0, 'escapechar': None, 'false_values': None, 'comment': None, 'compact_ints': False, 'delim_whitespace': False, 'thousands': None, 'dayfirst': False, 'verbose': False, 'nrows': None, 'encoding': None, 'quotechar': '"', 'dialect': None, 'warn_bad_lines': True, 'delimiter': None, 'compression': None, 'engine': 'python', 'error_bad_lines': True, 'keep_date_col': False, 'sep': '\t', 'iterator': False, 'doublequote': True, 'dtype': None}

--------------------- >> end captured stdout << ----------------------

----------------------------------------------------------------------
Ran 1 test in 0.002s

note the 'engine': 'python'.
and Issue #16013: Fix CSV Reader parsing issue with ending quote characters. looks suspicious.
is that enough to hit your fast-path for bug fixing?

edit:: wrong method but the conclusion is true, it's the python engine and not the c engine that the error comes from.

[ghost]

@juliantaylor , can you confirm the fix in bc6ad16?

[wesm]

Oh interesting. That makes sense that they altered the CSV module in Python 3.3

[juliantaylor]

the fix works, thanks

[ghost]

closed via bc6ad16 .
Test suite now passes on python 3.3